A solution to the single-question crowd wisdom problem
D. Prelec, H. Sebastian Seung, and J. McCoy, Nature (2017)
presented by K. Sasahara

Democratic methods of opinion aggregation

  • Majority vote
  • Weighted by confidence
  • Max confidence

Known issues

  • Weighting people equally
  • Weighting by confidence
  • These methods are biased for lowest common denominator information, at the expense of novel or specialized knowledge that is not widely shared.

US state capital question (see Fig. 1)

Question (P)

  • Is Philadelphia the capital of Pennsylvania?
  • No (i.e., Harrisburg)
  • Respondents voting yes believe that almost every one will agree with them, whereas respondents voting no expect to be in the minority.
  • Shared belief: Philadelphia is a big city. 

Question (C)

  • Is Columbia the capital of South Carolina?
  • Yes

A Solution

  • To utilize the knowledge of a well-informed subgroup in the cloud.

Surprisingly popular algorithm (see Fig. 2)

  • For a given question, people are asked two things:
  • What they think the right answer is (brief)
  • What they think popular opinion will be (reasoning)
  • Select the answer that is more popular than people predict
  • The biased coin model
  • Two world and a biased coin
  • Majority opinion favors yes in both worlds
  • People know the coins biases but don’t know which world is actual.
  • If predicted yes vote > actual yes vote, no is surprisingly popular.
  • In (P), actual vote 2/3 (0.67), prediction 5/7~6/7 (0.71~0.86); yes is less popular than predicted
  • In (Q), actual vote 2/3 (0.67), prediction 1/4~1/2 (0.25~0.5); yes is more popular than predicted
  • Identical distributions of confidence can arise in different settings.

Experiments (see Fig. 4, 5)

  • SP is the best performer across all studies
  • Study (2) by MTerk
  • In the art domain, the majority opinion is too conservative.
  • The majority voting is strongly biased against the high category
  • People have a belief that expensive pieces (> 30K) are rare.

Summary

  • Not only people’s actual beliefs but also their reasoning about the beliefs can be exploited to recover truth even when traditional voting methods fail.